Bayesian structural decomposition of streamflow time series

Abstract

Due to the significant influence of climate change and human activities on the water cycle, accurately estimating short- and long-term water availability has become imperative. This study introduces a time series model specifically crafted to decompose river flow time series, enabling estimation of trends, seasonality, and long memory components. This decomposition is interesting as it allows to separate permanent patterns, which can be associated with climate change processes, from transient effects on flow patterns. Additionally, this decomposition is incorporated into the quantile regression in quantile regression framework using a gamma function link. The estimation of this model is based on Bayesian inference, exploring the computational efficiency and accuracy of Integrated Nested Laplace Approximations. This methodology is applied to the principal rivers within the Araguaia River basin in Brazil and compared with other alternative time series decompositions with results indicating a remarkable alignment between the model and observed data.